Anthropic introduces invisible watermarking for Claude AI to meet EU transparency rules
Anthropic is adding invisible watermarks to text generated by Claude models, including future versions like Opus 5, to help identify AI‑generated content and comply with the EU AI Act.
Anthropic is embedding invisible watermarks into text produced by Claude models, including future versions like Opus 5, to satisfy the EU AI Act’s transparency requirements and help distinguish AI‑generated content from human‑written material.
Anthropic has begun embedding invisible watermarks in text generated by its Claude AI models, a step designed to bring the company's output into line with European Union AI Act transparency rules. The system can
How the watermarking system works and what it covers
Anthropic has built an invisible watermarking system for Claude that embeds subtle statistical patterns into every token it outputs. Those patterns can be scanned later to confirm whether a passage was produced by Claude, and the same technique can be applied to documents that have been proofread or edited by the model. By making the signal detectable without altering readability, the method satisfies the EU AI Act’s demand for clear provenance of synthetic text. The company says the watermark works across languages and can be toggled on or off for specific use cases, while remaining robust against minor rewriting or formatting changes. According to reports, the approach also helps differentiate AI‑assisted drafts from wholly human‑written content, supporting compliance and accountability in regulated markets.
| Functionality | Editability | Coverage |
|---|---|---|
| Anthropic detailed how its invisible text watermarking system for Claude models works to comply with EU AI Act transparency rules. | Anthropic shared details about the functionality, editability, and implications of the watermarking system for code. | The watermarking system is designed to identify content generated by Claude and can also mark human-written documents that were proofread or edited by Claude. |
| Not reported | Not reported | Not reported |
Revenue growth and market context
Anthropic reported revenue above $11.5 billion in the second quarter of 2026, underscoring the commercial scale of its AI operations as it rolls out compliance tools. The company introduced an invisible watermarking system for Claude AI to meet EU AI Act transparency rules, embedding identifiers into generated text to distinguish AI‑produced content and to mark documents edited by Claude. The watermark can also be applied to human‑written material that has been proofread or revised by the model. Some users have indicated they are cancelling subscriptions, fearing the watermark might trigger automated flags that question authorship. The latest model, Claude Opus 5, launched on July 24 2026, and the company has expanded Auto Mode in Claude Code for independent code work. The development reflects Anthropic’s effort to align its rapidly growing business with emerging regulatory expectations
User concerns over the watermark’s impact on authorship
Some subscribers have canceled their accounts, fearing that the watermark could trigger false accusations of AI authorship and raise legal questions about content ownership. Users worry the invisible marker might mislabel human-edited text as machine-generated, leading to disputes over copyright or plagiarism claims. This concern reflects broader unease about how automated attribution could affect freelance writers and editors who rely on clear provenance for their work. According to the outlet, the backlash highlights how compliance measures intended to clarify origins may inadvertently complicate them. The company has not publicly addressed the cancellations.
Future rollout and regulatory outlook
Anthropic plans to embed its invisible watermarking system into every new Claude model, including the recently released Opus 5, which launched on July 24, 2026. The rollout is intended to align with the EU AI Act’s transparency requirements and to extend the technology’s reach across the company’s product line. By integrating the watermark by default, Anthropic aims to help regulators verify the source of AI‑generated text and to flag documents that have been edited or proofread by Claude. The company will track regulatory feedback and user reactions as the feature spreads, adjusting implementation based on emerging compliance expectations. This phased approach reflects Anthropic’s broader strategy to embed governance tools within its AI offerings while maintaining technical performance.
Frequently asked questions
How does Anthropic's watermarking system work in Claude AI?
The system adds invisible watermarks to text generated by Claude models. It can also mark human-written documents edited by Claude. This helps identify AI-generated content and comply with EU AI Act transparency rules. The watermarking is designed to distinguish Claude output from human writing.
Why are some users canceling Claude subscriptions?
Some users canceled subscriptions after Anthropic introduced watermarking. They worry the invisible marks could cause clients to flag their work as AI-generated. This might raise questions about authorship and ownership of edited content.
What is Claude Opus 5 and when was it released?
Claude Opus 5 is the latest version of Claude, released on July 24, 2026. It is part of Anthropic's series of large language models. The model was launched as part of Anthropic's efforts to advance AI capabilities.
How does Claude's watermarking affect human-edited documents?
The watermarking system can mark human-written documents that were proofread or edited by Claude. This means even content partially created or revised by the AI may carry the invisible identifier.
Why is Anthropic adding watermarks to Claude AI text?
Anthropic is adding invisible watermarks to comply with EU AI regulations. The system helps establish whether text was generated by Claude or edited by it. This supports transparency requirements under the EU AI Act.
